Here's my “light-bulb” AI idea, (kinda out of left field)

  • 18 Replies
  • 613 Views
*

Hopefully Something

  • Roomba
  • *
  • 12
  • So where are these cookies?
First you need a container for all the gel. It shouldn’t get runny at high temperatures either, and it has to be optically clear and electrically unconducive.

What you do is you create artificial neurons and stir them into the gel, more or less.

The neurons are a sandwich of three parts; the top being a solar (photo-voltaic) wafer, the middle a silicon memory wafer, the bottom a wireless frequency emitter/receiver. (They might need a clear case to trap air for near neutral buoyancy in the gel.) With current technology (looking at the size of 200 GB micro SD cards, and researching the smallest commercial photo-voltaic cells) I imagine these things could be made the size of coarse sand, a few cubic millimeters.

LED’s coat the inside of the skull, powering all neurons without the need for billions of wires. Neurons communicate wirelessly with very low power signals.

To reduce the risk of overheating the gel should be thermally conductive. The neuron substitutes should be coated with something reflective to spread the light and prevent absorption (except the solar cells). Thirdly, the skull should be a double walled type thing with thermocouples and water cooling.

If a few billion of these artificial neurons are made, given identical programming which can change/activate on its own according to stimulus, and given human sensory inputs, well… hopefully they’ll organize themselves like biological neurons do. If we can get this to happen, the result could be something similar to biological intelligence.

Plus, if the designer does their job there will be cool glowy eyes. Just saying.

Comparing this to modern AI ventures in a metaphor; I think its simpler to make the right seed, and then feed it. It seems better than trying to make an adult tree from scratch. I mean, you probably could, and then get very repeatable trees. But there might be an easier way.

A fully formed brain might be fiendishly complex but it’s beginnings are much less so. The beginning of a human is the combination of two cells, each with 80 or so gigabytes of genetic information. Most of that is superfluous hiccup reflexes from our fish ancestors and such, or so I have gathered. After that life and intelligence develop by interacting with the environment.

Also, the body is important. It allows and more importantly, motivates the mind to do things. A body provides needs, things to get and get away from. If a baby didn’t have thirst, hunger, fear, and curiosity warring for control, then that baby wouldn’t understand the point of doing anything. Broad self driven-ness and initiative is what separates a mind and computer.

Maybe doing enough to meet the complex requirements needed to operate our bodies (and keep them comfortable) naturally gives rise to intelligence. Make one, watch its counterpart form.

So yeah, instead of putting knowledge into a structure, make a structure capable of acquiring knowledge.           

In summary, all we need to do is replicate the starting state of a neuron, make lots of copies, provide inputs, and let nature do it’s thing.


« Last Edit: June 28, 2018, 08:43:49 pm by Hopefully Something »

*

korrelan

  • Trusty Member
  • **********
  • Millennium Man
  • *
  • 1010
  • Look into my eyes! WOAH!
    • Google +
Welcome HS

Quote
hopefully they’ll organize themselves like biological neurons do

You were doing so well up until this point lol.

Perhaps just a black box with loads of blinky blue lights would suffice...

ED: Apologies... I'm being negative... keep us updated on your progress.

 :)
It thunk... therefore it is!

*

ranch vermin

  • Not much time left.
  • Terminator
  • *********
  • 790
  • Its nearly time!
Korrelan, just because he doesnt have much ANN experience doesnt mean his hardware idea isnt a good and factual one.

Getting rid of the wires is so important.
youll get 27 million nodes in a 30cm volume (unless my quick calc was wrong),  if your system doesnt overheat.
u think a millimetre is the only reduction you have to do for your cells?
but the amount of synapse record could be 10x10x10 in the cell,   and ull get 27 trillion synapses, then if I were you id split it over time communicating the 1000 memory bits - so it only needs 1 transmitter.

Youd need a tiny little oscillator as well,  and probably addresses to send the bits to,   so it probably gets a bit worse from there,  but probably not much more.

You couldnt call it solid state hardware :),  but it probably acts like it, and the cell itself would be I guess.



ED: i edded it a few times after i posted as well.
« Last Edit: June 27, 2018, 08:44:57 pm by ranch vermin »

*

korrelan

  • Trusty Member
  • **********
  • Millennium Man
  • *
  • 1010
  • Look into my eyes! WOAH!
    • Google +
I'd already updated my post lol.

He might be an AI guru for all I know, but unless you know of any current tech even remotely like he/ she is describing/ requiring then I will remain... cautious.



 :)

Ed: Actually what HS is describing is very similar to what I'm doing, except I'm using virtual space not a gel.

 :)
It thunk... therefore it is!

*

Art

  • At the end of the game, the King and Pawn go into the same box.
  • Global Moderator
  • ********************
  • Cleo
  • *
  • 4961
Welcome aboard HS! Good to have new people here.
Semantics...one person's vision might not necessarily speak for nor resemble those visions of anyone else. It's called, living the dream!

A bit of background might be a nice idea if you find the time and deem it necessary. We have a section for that too.
« Last Edit: June 28, 2018, 03:11:31 pm by Art »
In the world of AI, it's the thought that counts!

*

ranch vermin

  • Not much time left.
  • Terminator
  • *********
  • 790
  • Its nearly time!
I read HS Semantics, like it was a nick-name.  :)

*

Hopefully Something

  • Roomba
  • *
  • 12
  • So where are these cookies?
"hopefully they’ll organize themselves like biological neurons do"

Yep, I also found this to be a weak point. :) Thanks for replying though, I'm a 22 year old engineering student in college. It's not even electrical engineering, (it's mechanical), oh well. Still, a melding of two disciplines can introduce a whole new dimension.  I have some spare time to daydream but it'll be a while before I acquire some spare money to do physical experiments.

So with the neurons arranging themselves, (since the idea is to copy nature since that's a thing that I know works), I think the first thing to try is having them all "connected" at the beginning, and then prune away the unused connections, strengthen the useful ones, and so gradually build on what you got right by chance in the beginning.
To get them to do this, i'll try to replicate the behavior of a baby neuron as closely as possible with code. This is where I'm clueless, I mean I made some lights blink with an arduino uno once. Then again, the behavior of a single cell might just be simple enough to replicate.
So maybe cell behavior is the thing to research next.
 This has got me thinking about the whole thing again. Thanks cool people on forums!

*

korrelan

  • Trusty Member
  • **********
  • Millennium Man
  • *
  • 1010
  • Look into my eyes! WOAH!
    • Google +

*

ranch vermin

  • Not much time left.
  • Terminator
  • *********
  • 790
  • Its nearly time!
Excuse me for butting in on this cool idea ->

Thats cool Korrellan,  but the weight in a cell and the cell ids to synapse to have to be variable, and I dont think youd need anywhere near as much.

U should make his head a light bulb,  that would be funny.  Should keep things simple,  miniaturization can be over miniaturization,  I think all this reduction going on is scientists that know bugger all but it, and its completely superfluous for even advanced technology except uselessly showing off.

*

Art

  • At the end of the game, the King and Pawn go into the same box.
  • Global Moderator
  • ********************
  • Cleo
  • *
  • 4961
It is thanks to that miniaturization that we can now have such devices as smartphones, tablets, laptops, flat-screen TVs and all sorts of electronic wonders.
Remember those photos of the first computers in the 40's and 50's that filled an entire room. You now possess more than 1,000 times that power in your smartphone (if you have one). So yes, it might get like today's autos that almost require a college grad to work on them (with a computer, I might add), but the forward goal is to keep pushing the limits of smaller, faster, better with miniaturization and nanotechnology. It's our future unless one would rather hop in the horse-drawn buggy and slide the beads of an abacus. Just sayin'... ;)
In the world of AI, it's the thought that counts!

*

ranch vermin

  • Not much time left.
  • Terminator
  • *********
  • 790
  • Its nearly time!
How about a robo-horse pullin a buggy. 

Definitely less mess left behind.

*

Zero

  • Trusty Member
  • ********
  • Replicant
  • *
  • 648
    • Thinkbots are free
Yes, and it will probably end up looking like magic. We'll probably go back to earth, with simple life, only this little "plus", magic. Shape changing objects, light without electricity, ...etc.
Thinkbots are free, as in 'free will'.

*

ranch vermin

  • Not much time left.
  • Terminator
  • *********
  • 790
  • Its nearly time!
no thanks exactly, i like it when i can understand what im doing - technology all the way.

*

Zero

  • Trusty Member
  • ********
  • Replicant
  • *
  • 648
    • Thinkbots are free
I understand that. :)
Thinkbots are free, as in 'free will'.

*

Hopefully Something

  • Roomba
  • *
  • 12
  • So where are these cookies?
Re: Here's my “light-bulb” AI idea, (kinda out of left field)
« Reply #14 on: July 11, 2018, 08:18:37 pm »
Could you employ a genetic algorithm to get a rough organisation from these artificial neurons? An outside computer could wirelessly try different connections between them. If you connected a camera to the brain via an electrode panel, and gave it some possibilities in the shape of legs, also with electrodes, (one for each muscle or something like that), the program would eventually hone in on the basics and get a zombie/Frankenstein thing going. Once it got to a certain developmental point the AI would be self sustaining and the computer could be disconnected. Then the AI could keep improving its self because it could now generate it's own useful experiences by exploring the environment.
What basic things should the genetic algorithm teach it to get it going? How many nodes could an average desktop PC handle? Each neuron should be given it's own code/number/name so messages could be directed specifically and repeatably to the same places right?